RHIT Parallel Computing - Updates and Resources

• Pseudo-Random Number Generator Daemon ( 1 Nov 02)
• GSL Libraries (11-Jan-02)
• Install (22-Oct-01)

Pseudo-Random Number Generator Daemon

I've attached a couple of files that may be of use to you and Mr. Huey before we meet this afternoon. The files are:
1) the C source code for an application that obtains random numbers using the PRNGD daemon,
2) a BASH shell script for use by PBS (the batch queuing system on the cluster), and
3) a couple of output files generated by running the test_rand application 10 times using PBS.

We can discuss the content, principles, or anything else related to using the cluster. We'll also discuss the other software issues you mentioned in your email, but running MATLAB is problematic at this time. It's not because we don't have a Linux version, we do. The issue is that the compute nodes can't reach the license manager because they're on a private network, only Brain can reach the license manager at this time.

PBS and OpenSSL have several man pages you might be interested in reading. They are: pbs, qsub, qstat, RAND_egd, and RAND_bytes.

Attachments in this folder:

GSL Libraries (11-Jan-02)

Subject: New scientific library on Brain, the RH parallel computer
Date: 11 Jan 2002 14:38:57 -0500
From: Stephen Jones <Stephen.Jones@Rose-Hulman.edu>
To: Allen.Broughton@Rose-Hulman.edu, Michael.Cain@Rose-Hulman.edu, Joshua.Holden@Rose-Hulman.edu, Daniel.Jelski@Rose-Hulman.edu, Stephen.Jones@Rose-Hulman.edu, Jeffery.Leader@Rose-Hulman.edu, Luther.Wang@Rose-Hulman.edu, Mark.Yoder@Rose-Hulman.edu
CC: turcotte@Rose-Hulman.edu

Per a request by Professor Cain, I have installed the GNU Scientific Library (GSL) on the Rose parallel computer, Brain. The library was built using the GNU C compiler, gcc. The GSL header files, libraries, and man pages are located in /usr/local. The URL for the GSL is http://sources.redhat.com/gsl/. This site includes a great deal of information about the package. The reference mail is installed locally in /usr/local/gsl. There are PostScript and HTML versions available.
The PostScript version is named gsl-ref.ps. The URL for the local HTML version is "file:/usr/local/gsl/gsl-ref-html/gsl-ref_toc.html".Since this was compiled with the GNU C compiler the libraries and include files will almost certainly NOT be compatible with the Portland Group compilers that are included in the Cluster Development Kit.

Enjoy.
Steve

Install (22-Oct-01)

Subject: Access to the RHIT parallel computing cluster
Date: Mon, 22 Oct 2001 21:02:37 -0500
From: "Stephen Jones" <c.stephen.jones@rose-hulman.edu>
To: <holden@rose-hulman.edu>, <yoder@rose-hulman.edu>, <brought@rose-hulman.edu>, <jelski@rose-hulman.edu>,
<leader@rose-hulman.edu>, <cain@rose-hulman.edu>
CC: "HelpDesk" <helpdesk@rose-hulman.edu>, "System Admin" <systems@rose-hulman.edu>, "Louis Turcotte" <turcotte@rose-hulman.edu>

You have requested, and been granted, access to the Rose-Hulman Institute of Technology parallel computing cluster. The host name for this system is brain.rose-hulman.edu and its IP address is 137.112.8.88. Access to Brain is via the Secure Shell protocol. SecureCRT for Microsoft Windows is one client and ssh on Solaris or Linux is another client. No other login protocol is supported, including TELNET and RLOGIN. I have attached an introductory document that describes the cluster
and its nodes, the software available, documentation, and the location of example source code.

Please contact Barbara Myers at the HelpDesk for your password.

Stephen Jones

File: cluster.intro.txt
Date: October 22, 2001

This is the Rose-Hulman Institute of Technology parallel computing cluster. It is composed of one master node and 18 computational nodes. Each node has two 1-GHz Intel Pentium III processors, 1 GB of RAM, two Fast Ethernet (100 Mbps) network interfaces, and a hard disk drive. The node interconnect is a private Fast Ethernet connection; only the master node, Brain, has external connectivity. The operating system is Red Hat Linux 7.1.

The Portland Group's Cluster Development Kit (CDK) provides the development environment for parallel applications. There are no commercial parallel applications currently available for this system. The following software
development tools and libraries are available:

• pgcc - C compiler, supports OpenMP 1.0
• pgCC - C++ compiler, supports OpenMP 1.0
• pgf77 - Fortran 77 compiler, supports OpenMP 1.1
• pgf90 - FORTRAN 90 compiler, supports OpenMP 1.1
• pghpf - High Performance FORTRAN compiler
• pgprof - Performance profiler
• pgdbg - Debugger

Documentation in HTML format is available online at:

1. file:/master/pgi/doc/index.html (on a machine with an X-windows client, open up a ssh session and type netscape file:/master/pgi/doc/index.html at the command prompt
2. http://www.pgroup.com/docs.htm

Traditional "man" pages are also available, e.g. man pgf90.

Also available is:

• OpenPBS - Portable Batch System, a batch job queuing system
• MPICH - Open source Message Passing Interface (MPI)
• ATLAS - Automatically Tuned Linear Algebra Software

Example source code is located in /master/pgi/bench and /master/pgi/EXAMPLES. I recommend you look in the directory /master/pgi/bench/mpihello for a parallel FORTRAN version of the standard HelloWorld program. Compile and run it to verify your environment and to get a feel for the tools. This program can be run immediately by typing:

• mpirun -np 4 /master/clustertest/runtest
  

You can use any number of nodes (NP 4) up to a maximum of 19 (18 if you are trying to avoid using the head node for computation).

Please feel free to explore the system; at some point you may become the local parallel programming guru.